Loading Nifty 50, Nifty Bank, and Sensex…
Cross-sectional research

Factor Lab

Combine transparent factor lenses over persisted filings and completed daily prices. Missing evidence remains unavailable.

Research Intelligence
Choose at least three factors. Every score is an empirical percentile; it is not an analyst rating.
What this page is for
What the ranking actually is

Every score is an empirical percentile against listings measured on the same exchange: 80 means it scored above 80% of its measured peers on that factor. It is computed from persisted filings and prices, never from an analyst opinion.

Why calibration is here

Calibration asks whether past scores lined up with what followed. It freezes each score on its capture date, waits for the completed 20-session window, and groups those finished outcomes by score band. Captures whose window has not completed are counted as pending rather than estimated, which is why a newly covered exchange shows an empty table instead of a number.

Why diagnostics are here

Diagnostics test the ranking itself: whether score order matched outcome order, how long that held, whether the top and bottom quintiles separated, how much the list churns, and whether two factors are the same bet twice. Weak diagnostics mean the ranking describes the exchange rather than beating it.

Why the rank button greys out when the page opens

The ranking runs automatically when the page opens, so that button is disabled only while its one request is in flight and re-enables when the banner above updates. Calibration and diagnostics load only when you press their own buttons, and nothing else runs in the background.

What each number means
Frozen score — the score exactly as it stood on its capture date, never restated later, so calibration cannot be flattered by data that arrived afterwards.
Coverage — the share of a factor's underlying metrics actually present for that listing. The same score on thin coverage rests on fewer inputs.
Pending — captures whose forward window has not completed yet. They are counted and excluded rather than estimated, so an empty table means not yet, not no relationship.
Mean rank IC — the average rank correlation between score order and what followed. Near zero means the ranking carried no ordering information over that window.
Actual top-quintile turnover — how much of the ranked list changed from one session to the next. High turnover means acting on the ranking would have meant constant replacement.
Factor redundancy — the rank correlation between two factors on the same exchange. Two that correlate highly are one bet held twice, whatever their names suggest.
Choose factors and run the lab.
Historical benchmark calibration

Group frozen daily scores by completed 20-session excess return. This is historical calibration, not predictive accuracy.

Calibration appears after enough post-deployment observations mature.
Factor diagnostics

Descriptive post-capture evidence for this exact exchange. Statistics are not a forecast or recommendation.

Choose a lookback and load persisted diagnostics.